Deep Reinforcement Learning-Based Optimal Scheduling for Multi-Energy Virtual Power Plant
Multi-energy virtual power plant(MEVPP)aggregates various forms of distributed energy and demand-side flexibility resources including electric and thermal energy.In order to realize the optimal scheduling of MEVPP,in this paper,a MEVPP model including power generation units,heating units,energy storage unit,air conditioning loads cluster,and demand response loads cluster is established,a deep reinforcement learning(DRL)-based optimal scheduling method is proposed for this model,corresponding reward function,state and action spaces are designed.The method is based on the proximal policy optimization(PPO)algorithm,which can regulate distributed energy and demand-side flexibility resources based on environmental information such as forecasted load,wind/light output,outdoor temperature,etc.,and obtain the strategy set of MEVPP optimal scheduling with the objective of minimizing the operating cost.The case study result proves the feasibility of DRL in MEVPP optimal scheduling and the extensibility of the strategy set.
virtual power plantmulti-energy complementaryoptimal schedulingdeep reinforcement learning
NETL
NSTL
万方数据
